Facilitating subgroup identification: the use of decision trees in breast cancer screening uptake

Abstract Introduction Organized breast cancer screening (BCS) programs are effective prevention measures for women aged 50-69 to prevent the sixth cause of death in Germany. It is unclear which sociodemographic factors and potential interactions are associated with never-attendance. Explorative approaches such as decision trees offer an inductive approach to intersectional subgroup identification based on different computational decisions. This study discusses two approaches for estimating the likelihood of not attending BCS in Germany. Methods Women aged 50-69 (N = 3,644) were selected from the German 2020 European Health Interview Survey. We estimate the risk of not attending BCS first by employing cross-classification regression analysis utilising known social dimensions and second by leveraging the nodes of the best-performing decision tree as exposure variables. Results The first approach, adjusting for age, found that women born in Germany had 1.97 higher odds (p = 0.015) of never attending BCS than high-income women not born in Germany. Conditional Inference Tree (mincriterion= 0.6040046, maxdepth=3) adjusted to optimise sensitivity and the Area Under the Curve (71.43% sensitivity, 63.21% AUC), identified 8 relevant intersectional groups characterised by unique combinations of family status, perceived social support, working situation, experienced physical limitation, and region, with the highest odds of 4.85 (p < 0.001) of not attending BCS for women living with their couple and children or non-conventional household constellations who perceived little, uncertain or a lot of social support. Conclusions Using explorative decision trees for subgroup identification, the study provides more detailed insights into women at higher risk of not attending BCS in Germany, aiding, on the one hand, more precise health reporting data and, on the other hand, enhanced preventive public health interventions. Key messages • Explorative decision trees for subgroup identification allow for more detailed insights into women at higher risk of not attending BCS in Germany for enhanced preventive public health interventions. • Women living with their couple and children or non-conventional constellations who perceived little, uncertain or a lot of social support are at the highest odds of never attending BCS in Germany.